On Databases and Data platforms for the Edge/Iot. Q&A with David Rolfe

Q1. The database/ data platform market is undergoing a consolidation. What is your take on this?

A number of factors are working together. At one point the 451 Group was tracking around 130 separate database / data platform products. We’re now seeing steady shrinkage. Some of this is the natural operation of the free market. Literally every possible idea was tried; some worked and some didn’t. The market simply isn’t big enough to support 130 platforms, be they open source or otherwise. 

But other factors are in play as well. The hyperscalers have also derailed the plans of companies that used VC funding to ‘launch’ an open source NoSQL platform. The original assumption was that enterprise support contracts from rich Fortune 500 companies would allow a commercial operation to succeed, while simultaneously nurturing an open source version. That changed the day hyperscalers started charging money for hosted versions of open source products. This, in turn, led to the Server Side Public Licence, which aims to prevent hosting plus reselling to the public. The recent shift to DBaaS is the latest move in this chess game.

As to where it’s all going? On-prem isn’t going away. Beyond that, it’s hard to predict with certainty.

Q2. Considering high cloud bills and that many Edge/Iot use cases have what might be called ‘finely balanced’ economics. Are we approaching ‘peak cloud’?  

I think we’re at ‘peak cloud’. We’re getting a lot of interest as the long-term economics of cloud ownership become visible to the corner office. Take one example: One of the big advantages of cloud is that if you want to turn a three-node cluster into a five-node cluster you can just click your fingers and add two new instances. Or can you? Because unless you reserve them in advance there is no guarantee that when you ask for them you’ll actually get them. Then there’s the perverse incentives at play. If I own hardware and my workload starts exceeding what I can do, I can use my in-house DB skills to optimize the application so it fits. If I’m using DBaaS, the vendor will provide support, but their incentive is to get me to use more by-the-hour resources, not less. 

Q3. ARM seems the favorite architecture for Edge/IoT. What is it? And why is it important?

ARM chips have been around for years in things like mobile phones but are now entering the server market. As a rule of thumb, an ARM chip is 40% faster than an x86 one, as they usually don’t have hardware multithreading. Edge likes ARM because it uses a lot less power than Intel and can be very affordable.

Q4. Why did you start doing TCO benchmarks on AWS Graviton, Intel C7 and even Raspberry Pi?

As a company, we’ve invested significant ‘blood and treasure’ on porting to ARM because our customers were asking us for that. Because Volt is a ‘close-to-the-metal’ product, this wasn’t a simple task, but we did it and got results. Pi was picked as a defined ‘minimum configuration’. Note that we don’t approve it formally for production as the memory cards can wear out, but a lot would depend on the use case, so that might change.

Q5. What did you discover in doing such TCO benchmarks?

Using our charging benchmark, AWS Graviton is 40% faster than AWS c5 and 20% faster than c7. Pricing is also very competitive, but since AWS could change it at any time it’s not what I’d ‘hang my hat on’ on for this one. 

As for the Raspberry Pi benchmark, when we started testing ARM I asked our engineers what the smallest hardware we could run on would be, because a lot of Edge hardware is … austere. Nobody actually knew, so I made it my business to find out. For under Euro 1000 I was able to build a five-node cluster and get around 2,500 TPS on our benchmark. Assuming the hardware was written off after a year, the cost per transaction was, bizarrely enough, about the same as AWS. 

Fifteen years ago, standing up a system that could support a workload of 2,000 TPS for an OLTP application would be a big deal. Now we can do it with the kind of hardware people do school projects on. The world is changing, and ARM is part of that change.

Q6. You recently published the Volt’s latest Yahoo! Cloud Serving Benchmark (YCSB) numbers. What did you discover this time? 

This is the first time we’ve started to closely track TCO as well as actual results. Getting to 1,000,000 TPS in YCSB isn’t helpful if it costs you US$1,000,000 to do it. What we’re seeing is that we can get 500K TPS for about US$2.50 an hour. We think we can go much higher, but at 500K TPS we’re starting to saturate the network available to us in AWS. We could go bigger if we set up a custom network, which we may try later this year.

Q7. You announced your latest release, Volt Active Data 13.1. What is special about it?

Many individual things:

• In-service upgrades: ie, the ability to upgrade the software on a Volt cluster while it’s running.
• ARM64 CPU support, which we’ve discussed above.
• Elastic reduction of Kubernetes cluster sizes. Elastic shrink is very, very hard to do in Kubernetes.

The big promise of ‘cloud native’ is that a lot of the stuff we used to regard as scary and time-consuming is automated and not a problem anymore. But when it comes to elastic add and shrink, many aspects of how K8tes works make it very hard to implement. Both of these features were hard to implement, but we did it.

Q8. What’s next?

We’re continuing research into an Edge-specific offering and an API offering. 

There are some interesting ideas out there about having a local database that’s mirrored to a cloud, the same way Apple’s iCloud does for your photos. This makes Edge activity visible centrally and also makes it really easy to replace the Edge system if it fails.

On the API front, we’re looking at a way to make Volt available as a web service with zero client coding. 

Resources

VOLT APPROACHES 500K TPS FOR <$2.50 PER HOUR ON YCSB

WHY V13.1 IS OUR MOST IMPORTANT RELEASE YET January 16, 2024

………………………..

David Rolfe brings 20+ years of experience managing data in the telecom industry. David helps telecom software vendors meet the scale and latency requirements imposed by 5G data utilizing Volt Active Data. He helps companies take the steps they need to deploy mass-scale, ultra-low latency
transactional applications in cloud-native environments. He has over 25 years of experience with high-performance databases and telco systems and demonstrated expertise with charging and policy systems. He has authored multiple patents relating to geo-replicated conflict resolution.

Sponsored by Volt Active Data.